· 4 min read
A few weeks ago, the Stanford Institute for Human-Centered Artificial Intelligence (HAI) released its annual AI Index—a report that aims to quantify the current state of AI across industries and the world.
Now in its fifth year, the report has added a new global survey of robotics researchers, data on AI policy in 25 countries, and a chapter on AI ethics.
The big number: Global private investment in AI surged to $93.5 billion, more than 2x that of 2020. But at the same time, fewer new companies are being created: Over the past three years, the number of newly-funded AI companies dropped from 1,051 to 762 to 746, per the report.
“It’s becoming clearer and clearer that not only is investment going up in AI, but it’s going mostly to more mature firms. It’s going in bigger chunks to more mature firms, which I take as an indication that…applications of AI [are] more likely to see their way into real use on a significant scale,” Ray Perrault, one of the authors of the AI Index, told us.
- But, Perrault added, there’s a downside: “It’s not clear from these figures how much easier it is for innovation that comes out of small startups to get funded.”
Around the world
Key countries: HAI released a Global AI Vibrancy Tool, which uses data from 2017–2021 to determine how countries stack up in the AI race. The US came out on top, followed by China and then India.
- The US ranked highest on metrics like newly funded AI companies, private investment, and patent grants, while China ranked higher in areas like patent applications and conference and journal publications.
- US-China collaboration on AI publications was at an all-time high last year.
In the legislatures of the world: In 2016, just one bill containing the phrase “artificial intelligence” passed; in 2021, that number increased to 18, according to the AI Index’s analysis of legislation in 25 countries. Last year, the US, the UK, and Spain each passed three pieces of AI-related legislation.
In the lab
Cheaper, faster, better?: The cost to train an AI image-classification system has plummeted almost 64% since 2018, while the training time for such a system has sped up by nearly 95%. The cheaper and faster trend also applies to other AI areas like object detection, recommendation, and language processing.
- Robotic arms have also gone on sale, with the median price falling more than 45% in five years. As of last year, one of the contraptions will set you back “just” $22,600—compared to $42k in 2017.
Stay up to date on tech
Drones, automation, AI, and more. The technologies that will shape the future of business, all in one newsletter.
Morality clause: AI ethics research is on the up-and-up. Since 2014, publications related to AI fairness and transparency have increased 5x at “ethics-related conferences,” the researchers wrote.
Another of the Index’s key points? Large language models, which are“setting new records on technical benchmarks,” per the researchers, are also more capable of reflecting biases learned from training data. For example, the researchers wrote, a 280-billion-parameter model that was developed last year showed a nearly 30% increase in “elicited toxicity” when compared to a 117-million-parameter model that was “considered the state of the art” in 2018.
- But corporate efforts to address ethical concerns may be lacking, according to one survey cited in the Index: “While 29% and 41% of respondents recognize ‘equity and fairness’ and ‘explainability’ as risks while adopting AI, only 19% and 27% are taking steps to mitigate those risks,” the researchers wrote.
Bottom line: We’re seeing the AI industry mature. Not only is it consolidating—from a VC perspective, at least—but machine learning models are also becoming cheaper, faster, and more accessible. That’s good news for budget-minded companies, but it’s likely not so good for preventing the potential harms that can result from the use of AI.